In India, Rail mode of transport serves as frequently preferrable transit systems operating with the optimal cost. Typically, the Indian Railways transport thousands of people on a day-to-day basis in addition to transporting large consignment of goods. Therefore, it is important for the trains to ensure that they run on quality tracks. At times, these tracks are challenged by the friction generated by continuous passage of trains in addition to over corrosions that occur due to their environmental imbalance. Preventive Track Maintenances (PTMs) have been recently introduced by railways for enhancing the quality of railway tracks, but on the contrary, it was failed to focus on the actual needs or emergencies of railway tracks. Moreover, none of the existing methods have been tested with real time datasets. Specifically, holding only two class labels are being considered resulting in the reduction of classification performances. But the major challenging task is that the real-time datasets fall under the category of multi-variant data. Hence, this study aims to provide a Decision Support System (DSS) that predicts the Railway Track Quality (RTQ) from the real time datasets available on the track inspection data of the Indian metro rail system. The proposed research uses clustering and classification processes for achieving Predictive Track maintenance (PTM). Furthermore, the proposed method of RPTMs includes five steps namely data collection, data transformations, clustering of data, preventive maintenances, and evaluations. The undertaken datasets are transformed into numeric formats for the creation of clusters using Kernel Mean Weight Fuzzy Local Information C Means (KMWFLICMs). The resultant clusters from the data have five major types of clusters such as Normal, Low risks, Medium, High, and Emergency Risks based on the parameters of gauge, cross level attributes, turnouts and versine of mainline. From the inferred cluster results, the dataset was further classified to choose maintenance status from four major classes namely No Actions, Fixed Maintenances, Investigate Maintenances, and Emergency Maintenance pertinent to the outcomes of FWCNNs (Fuzzy Weight Convolution Neural Networks). The proposed system was experimented on MATLAB and evaluated against various machine learning approaches. Therefore, the obtained statistical results confirmed that the proposed FWCNN model had afforded higher accuracy in predicting the maintenance interventions based on relevant risk category.
Read full abstract